Research On Variable Angle Of Robot Obstacle Detection Methods Based On Contour Recongnition

In the process of moving, intelligent robots need to perceive the obstacles around it. As a result, how to realize the real-time accurate obstacle detection becomes a hot topic in the ro-bot autonomic navigation field. By shooting the same scene simultaneously, we can obtain the depth information according to the triangulation principle and reconstruct the 3D shape of the scene. Since binocular stereo technology is informative and has a strong capability of envi-ronment perception, adopting binocular stereo vision to detect obstacles is significant.However, the visual angle of binocular stereo vision is small, so they can’t Perceive the environment information fully. Thus, the path planning on it is not the best, what’s more, maybe it’s wrong. Realizing detecting obstacle in variable angle according to the practical re-quirements is urgent.Focusing on how to realize detecting obstacle in variable angle by binocular stereo vi-sion, the article has done the following works:(1)Introduced the camera keyhole model, camera intrinsic parameter model and outside parameter model, camera calibrationprinciple, some common camera calibration methods and binocular camera stereo calibration in detail. In this paper, the calibration method of Zhengyou Zhang is used for binocular camera calibration.(2)Edge feature points detection was carried out thorough research. Because of the vari-ous influence in a real environment, the measured edge was always discontinuous, so we use close calculations to connect the breaking edges.(3)Based on the comparison of different stereo matching algorithms, we use win-dow-based gray matching principle to match the pixels on contour. The experiment prove that the algorithm not only calculates quickly, but also has high rate of accuracy.(4)In order to reduce the calculation amount effectively and retain the information of ob-stacles on the greatest degree, the article raised a 3D-reconstruction method based on object contours in picture. Firstly, we should match the contours extracted from left image and right image. Secondly, window-based gray matching principle will be used to match the pixels on the matching contours. And then we can regain the 3-D information of the obstacle contours. What’s more, we can optimize 3-D point cloud based on the continuity of contour. In order to evaluate the algorithm, we detect some obstacles in a real environment. The experiment prove that the 3D-reconstruction method based on object contours in picture can get accurate 3-D coordinate of object contours and satisfy real-time requirements in robot navigation.(5)The small view of binocular stereo vision result in the failure of detecting obstacle entirely. So, we raised a 3-D mosaic algorithm based on dual registration within the variable view. Firstly, match the contours in left images P11, P12 of two sets to find the overlapping region of two 3-D point cloud. Secondly, calculate the center of gravity of two 3-D point cloud and extract features, then, use Seven parameter method to calculate the rotation matrix Ri and translation matrix Ti of two sets of 3-D point cloud, thus, the intial registration was done. Thirdly, it weighs the corresponding points of the overlapping region. And then, calcu-late based on iterative closest point method to get the rotation matrix Ri. and translation ma-trix Ti of two sets of 3-D point cloud. thus, the point cloud is accurately registered. Fourthly, Use the weighted average fusion algorithm to fuse the 3D point cloud after registration. So we can defects such as Eliminate splicing cracks and get the clear and complete 3-D point cloud. The experiment results indicate that the method proposed in this paper is accuracy, fast and less time consuming.